| name | gather-context |
| description | Research first SOP before implementing any code change. Use when starting any task that involves modifying an existing codebase features bugs refactors open source contributions. Triggers when I want to work on X help me implement or fix or refactor X gather relevant context for X lets work on this. Launches parallel atlas subagents for local codebase evidence plus voyager external signal research, then presents 3 ranked approaches minimal diff first for user approval before any code is written. |
Gather Context
Research a codebase before touching it. Parallel research -> synthesis -> 3 ranked approaches -> wait for approval.
Primary goal for open source: Changes must look like the maintainer wrote them. Minimal diff. Maximum style alignment.
Phase 0 — Issue Intake
Before research, frame the task from the issue/request itself.
Artifact directory rule:
-
gather-context creates ISSUE_DIR before research starts.
-
ISSUE_DIR = _ai/task/{YYYY-MM-DD}/{slug} using the create-issue-compatible nested shape.
-
Reuse the same ISSUE_DIR for every artifact in this run.
-
Restate the issue in plain language
-
Extract explicit acceptance criteria
-
Mark assumptions and missing product decisions
-
Classify the task: bug | feature | refactor | claim-check
-
Create or update {ISSUE_DIR}/issue.md as the proposal source of truth
issue.md must contain these sections:
- Original GitHub Issue
- Acceptance Criteria
- Gherkin Happy Path
- Gherkin Edge Path
- Research Index
- Approaches
- Judge Decision placeholder
If the issue is underspecified in a way that would materially change the implementation, stop here and ask 1-3 targeted questions instead of forcing options.
Phase 1 — Define Target Scenarios, Then Launch Research Agents in Parallel
Before launching subagents, generate two Gherkin scenarios from the original user query: one happy path and one edge path. Keep them minimal and targeted. We are defining the smallest user-visible contract for a simple enhancement, not a full spec.
Use this format:
Happy Path: [User action and outcome]
Given [user state/precondition]
When [user action]
Then [user-visible outcome with verifiable condition]
Edge Path: [Boundary/failure case and outcome]
Given [edge precondition]
When [user action]
Then [safe user-visible outcome with verifiable condition]
Acceptance Criteria:
- [Measurable outcome: specific value/threshold/state]
Rules:
- Generate exactly 2 scenarios: happy path + edge path
- Base it on the original user query, not on implementation guesses
- Keep it user-visible, testable, falsifiable, and implementation-agnostic
- Keep it minimal and targeted to the enhancement being requested
- These scenarios are the target goal for all subagents
Spawn all five agents simultaneously using the Task tool:
- Agents 1-4 use
subagent_type: atlas
- Agent 5 uses
subagent_type: voyager
Pass both scenarios to every subagent as part of its task context so research stays anchored to the same target behavior.
Evidence requirement: Agents 1-4 must cite findings with code snippets, file paths, and line numbers. Agent 5 must cite URLs and quote/snippet the relevant source. No assertions without evidence. Each agent must save its full evidence report under {ISSUE_DIR}/research/. If results are thin or inconclusive, note gaps explicitly in Phase 2 — do not proceed with assumptions.
Agent 1: Code Archaeology
What does this code do today, and how?
Save full report to {ISSUE_DIR}/research/code-archaeology.md.
- Locate entry points, relevant files, core logic
- Trace the current implementation end-to-end
- Identify existing tests covering this area
- Note any TODOs, FIXMEs, or known issues near the target
Agent 2: Dependency Map
What breaks if we touch this?
Save full report to {ISSUE_DIR}/research/dependency-map.md.
- Map callers and callback/event consumers (what depends on this code, including who reacts to emitted/invoked behavior)
- Map callees and callback/event producers (what this code depends on, including what it emits/invokes)
- Capture invocation cardinality and order for critical interactions (once vs multiple, before vs after)
- Identify public API surface vs internal details
- Assess blast radius: files, modules, tests, and cross-component behavioral side effects at risk
- Flag any breaking change risks
Agent 3: UX Behavior
What does the user see today, and what will they see after?
Save full report to {ISSUE_DIR}/research/ux-behavior.md.
Given the feature/change being investigated, trace the user-facing path — not the code path.
- Current UX — step by step what the user sees and can do today (screens, states, limits, caps, hidden elements)
- Post-change UX — same walkthrough after the proposed change; explicitly call out anything that stays blocked, hidden, or broken
The main agent must tell this agent what feature is being added/changed so it can focus the trace. Output must be two clearly labelled sections: Current UX and Post-change UX. No prose beyond what is needed to describe user-visible behavior. Flag any gap where the post-change UX does not match user expectations.
Agent 4: Style Fingerprint
How does this codebase write code?
Save full report to {ISSUE_DIR}/research/style-fingerprint.md.
Start with files adjacent to the task. Expand repo-wide when touching shared infrastructure.
Look for:
- Naming conventions (variables, functions, types, files)
- Error handling patterns (exceptions vs return values vs Result types)
- How similar problems are solved elsewhere in the codebase
- Test structure and naming
- Code organization within files (imports, grouping, ordering)
- Linting/formatting config (
.eslintrc, pyproject.toml, .editorconfig, etc.)
- Commit/PR style if relevant (small focused changes vs large PRs)
Output: A concise style cheatsheet — bullet points only, no prose.
Agent 5: External Signal
What does the outside world say that should shape this implementation?
Save full report to {ISSUE_DIR}/research/external-signal.md.
Research platform conventions, framework behavior, API contracts, ecosystem norms, security/privacy guidance, accessibility rules, and unfamiliar implementation patterns relevant to the target scenarios.
Research priority:
- Official docs, specs, and platform guides
- Upstream repos, changelogs, maintainer issues, and discussions
- Reputable ecosystem examples from known teams or mature libraries
- High-signal articles only when they add concrete implementation guidance
Reject:
- SEO blogs
- Generic tutorials
- Uncited claims
- Stale guidance unless clearly labelled
Output:
- 3-7 bullets max
- URL citation per claim
- Version/date/platform notes where relevant
- What this changes about our approach
Phase 2 — Synthesize
After all 5 agents return, combine findings:
- Gherkin Happy Path — the happy path scenario from Phase 1, surfaced verbatim
- Gherkin Edge Path — the edge path scenario from Phase 1, surfaced verbatim
- Current UX — what the user sees and can do today (from Agent 3, surfaced here verbatim)
- Post-change UX — what the user will see after the change (from Agent 3, surfaced here verbatim)
- Current behavior — what the code does today
- Constraints — what must not change (public API, test contracts, style rules)
- Style rules — the extracted cheatsheet from Agent 4
- Blast radius — scope of impact from Agent 2
- External signal — cited implementation-shaping findings from Agent 5
- Research Index — links to all five files in
{ISSUE_DIR}/research/
Update issue.md with the synthesized scenario sections and Research Index. Keep full evidence in the research files, not in issue.md.
Decision Heuristics (Apply to every proposal)
Use these as hard filters before presenting options:
- Reuse first (DRY): prefer existing modules/components/patterns over new ones.
- KISS: choose the least complex approach that meets requirements.
- YAGNI: do not add extensibility/abstractions unless current requirement needs it.
- Single source of truth: avoid duplicated state/data paths.
- User trust/safety: no risky shortcuts that could create silent bad outcomes.
Phase 3 — Present 3 Approaches
Rank by: minimal diff + style alignment first -> more involved last.
Each option must include one regression probe describing how to verify no duplicate trigger/clobber regressions were introduced.
For each option ask: "Would a maintainer approve this PR without asking for changes?"
Reason from first principles: work backwards from the goal — what is the simplest change that satisfies the requirement without introducing concepts the codebase doesn't already use?
Write the 3 synthesized approaches into the Approaches section of {ISSUE_DIR}/issue.md. Do not create approaches.md, decision.md, handoff.md, plans/, or suggestions.md for this pipeline.
Phase 4 — Hard Stop
Present the 3 options and wait for explicit user selection.
Do not begin implementation until the user picks an approach.
